quota: rename quota functions from upper case, make bigger ones non-inline
[safe/jmp/linux-2.6] / fs / dquot.c
1 /*
2  * Implementation of the diskquota system for the LINUX operating system. QUOTA
3  * is implemented using the BSD system call interface as the means of
4  * communication with the user level. This file contains the generic routines
5  * called by the different filesystems on allocation of an inode or block.
6  * These routines take care of the administration needed to have a consistent
7  * diskquota tracking system. The ideas of both user and group quotas are based
8  * on the Melbourne quota system as used on BSD derived systems. The internal
9  * implementation is based on one of the several variants of the LINUX
10  * inode-subsystem with added complexity of the diskquota system.
11  * 
12  * Version: $Id: dquot.c,v 6.3 1996/11/17 18:35:34 mvw Exp mvw $
13  * 
14  * Author:      Marco van Wieringen <mvw@planets.elm.net>
15  *
16  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
17  *
18  *              Revised list management to avoid races
19  *              -- Bill Hawes, <whawes@star.net>, 9/98
20  *
21  *              Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
22  *              As the consequence the locking was moved from dquot_decr_...(),
23  *              dquot_incr_...() to calling functions.
24  *              invalidate_dquots() now writes modified dquots.
25  *              Serialized quota_off() and quota_on() for mount point.
26  *              Fixed a few bugs in grow_dquots().
27  *              Fixed deadlock in write_dquot() - we no longer account quotas on
28  *              quota files
29  *              remove_dquot_ref() moved to inode.c - it now traverses through inodes
30  *              add_dquot_ref() restarts after blocking
31  *              Added check for bogus uid and fixed check for group in quotactl.
32  *              Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
33  *
34  *              Used struct list_head instead of own list struct
35  *              Invalidation of referenced dquots is no longer possible
36  *              Improved free_dquots list management
37  *              Quota and i_blocks are now updated in one place to avoid races
38  *              Warnings are now delayed so we won't block in critical section
39  *              Write updated not to require dquot lock
40  *              Jan Kara, <jack@suse.cz>, 9/2000
41  *
42  *              Added dynamic quota structure allocation
43  *              Jan Kara <jack@suse.cz> 12/2000
44  *
45  *              Rewritten quota interface. Implemented new quota format and
46  *              formats registering.
47  *              Jan Kara, <jack@suse.cz>, 2001,2002
48  *
49  *              New SMP locking.
50  *              Jan Kara, <jack@suse.cz>, 10/2002
51  *
52  *              Added journalled quota support, fix lock inversion problems
53  *              Jan Kara, <jack@suse.cz>, 2003,2004
54  *
55  * (C) Copyright 1994 - 1997 Marco van Wieringen 
56  */
57
58 #include <linux/errno.h>
59 #include <linux/kernel.h>
60 #include <linux/fs.h>
61 #include <linux/mount.h>
62 #include <linux/mm.h>
63 #include <linux/time.h>
64 #include <linux/types.h>
65 #include <linux/string.h>
66 #include <linux/fcntl.h>
67 #include <linux/stat.h>
68 #include <linux/tty.h>
69 #include <linux/file.h>
70 #include <linux/slab.h>
71 #include <linux/sysctl.h>
72 #include <linux/init.h>
73 #include <linux/module.h>
74 #include <linux/proc_fs.h>
75 #include <linux/security.h>
76 #include <linux/kmod.h>
77 #include <linux/namei.h>
78 #include <linux/buffer_head.h>
79 #include <linux/capability.h>
80 #include <linux/quotaops.h>
81 #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
82 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
83 #include <net/netlink.h>
84 #include <net/genetlink.h>
85 #endif
86
87 #include <asm/uaccess.h>
88
89 #define __DQUOT_PARANOIA
90
91 /*
92  * There are two quota SMP locks. dq_list_lock protects all lists with quotas
93  * and quota formats and also dqstats structure containing statistics about the
94  * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
95  * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
96  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
97  * in inode_add_bytes() and inode_sub_bytes().
98  *
99  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
100  *
101  * Note that some things (eg. sb pointer, type, id) doesn't change during
102  * the life of the dquot structure and so needn't to be protected by a lock
103  *
104  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
105  * operation is just reading pointers from inode (or not using them at all) the
106  * read lock is enough. If pointers are altered function must hold write lock
107  * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
108  * for altering the flag i_mutex is also needed).  If operation is holding
109  * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
110  * dqonoff_mutex.
111  * This locking assures that:
112  *   a) update/access to dquot pointers in inode is serialized
113  *   b) everyone is guarded against invalidate_dquots()
114  *
115  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
116  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
117  * Currently dquot is locked only when it is being read to memory (or space for
118  * it is being allocated) on the first dqget() and when it is being released on
119  * the last dqput(). The allocation and release oparations are serialized by
120  * the dq_lock and by checking the use count in dquot_release().  Write
121  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
122  * spinlock to internal buffers before writing.
123  *
124  * Lock ordering (including related VFS locks) is the following:
125  *   i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
126  *   dqio_mutex
127  * i_mutex on quota files is special (it's below dqio_mutex)
128  */
129
130 static DEFINE_SPINLOCK(dq_list_lock);
131 DEFINE_SPINLOCK(dq_data_lock);
132
133 static char *quotatypes[] = INITQFNAMES;
134 static struct quota_format_type *quota_formats; /* List of registered formats */
135 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
136
137 /* SLAB cache for dquot structures */
138 static struct kmem_cache *dquot_cachep;
139
140 int register_quota_format(struct quota_format_type *fmt)
141 {
142         spin_lock(&dq_list_lock);
143         fmt->qf_next = quota_formats;
144         quota_formats = fmt;
145         spin_unlock(&dq_list_lock);
146         return 0;
147 }
148
149 void unregister_quota_format(struct quota_format_type *fmt)
150 {
151         struct quota_format_type **actqf;
152
153         spin_lock(&dq_list_lock);
154         for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
155         if (*actqf)
156                 *actqf = (*actqf)->qf_next;
157         spin_unlock(&dq_list_lock);
158 }
159
160 static struct quota_format_type *find_quota_format(int id)
161 {
162         struct quota_format_type *actqf;
163
164         spin_lock(&dq_list_lock);
165         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
166         if (!actqf || !try_module_get(actqf->qf_owner)) {
167                 int qm;
168
169                 spin_unlock(&dq_list_lock);
170                 
171                 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
172                 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
173                         return NULL;
174
175                 spin_lock(&dq_list_lock);
176                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
177                 if (actqf && !try_module_get(actqf->qf_owner))
178                         actqf = NULL;
179         }
180         spin_unlock(&dq_list_lock);
181         return actqf;
182 }
183
184 static void put_quota_format(struct quota_format_type *fmt)
185 {
186         module_put(fmt->qf_owner);
187 }
188
189 /*
190  * Dquot List Management:
191  * The quota code uses three lists for dquot management: the inuse_list,
192  * free_dquots, and dquot_hash[] array. A single dquot structure may be
193  * on all three lists, depending on its current state.
194  *
195  * All dquots are placed to the end of inuse_list when first created, and this
196  * list is used for invalidate operation, which must look at every dquot.
197  *
198  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
199  * and this list is searched whenever we need an available dquot.  Dquots are
200  * removed from the list as soon as they are used again, and
201  * dqstats.free_dquots gives the number of dquots on the list. When
202  * dquot is invalidated it's completely released from memory.
203  *
204  * Dquots with a specific identity (device, type and id) are placed on
205  * one of the dquot_hash[] hash chains. The provides an efficient search
206  * mechanism to locate a specific dquot.
207  */
208
209 static LIST_HEAD(inuse_list);
210 static LIST_HEAD(free_dquots);
211 static unsigned int dq_hash_bits, dq_hash_mask;
212 static struct hlist_head *dquot_hash;
213
214 struct dqstats dqstats;
215
216 static void dqput(struct dquot *dquot);
217
218 static inline unsigned int
219 hashfn(const struct super_block *sb, unsigned int id, int type)
220 {
221         unsigned long tmp;
222
223         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
224         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
225 }
226
227 /*
228  * Following list functions expect dq_list_lock to be held
229  */
230 static inline void insert_dquot_hash(struct dquot *dquot)
231 {
232         struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
233         hlist_add_head(&dquot->dq_hash, head);
234 }
235
236 static inline void remove_dquot_hash(struct dquot *dquot)
237 {
238         hlist_del_init(&dquot->dq_hash);
239 }
240
241 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
242 {
243         struct hlist_node *node;
244         struct dquot *dquot;
245
246         hlist_for_each (node, dquot_hash+hashent) {
247                 dquot = hlist_entry(node, struct dquot, dq_hash);
248                 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
249                         return dquot;
250         }
251         return NODQUOT;
252 }
253
254 /* Add a dquot to the tail of the free list */
255 static inline void put_dquot_last(struct dquot *dquot)
256 {
257         list_add_tail(&dquot->dq_free, &free_dquots);
258         dqstats.free_dquots++;
259 }
260
261 static inline void remove_free_dquot(struct dquot *dquot)
262 {
263         if (list_empty(&dquot->dq_free))
264                 return;
265         list_del_init(&dquot->dq_free);
266         dqstats.free_dquots--;
267 }
268
269 static inline void put_inuse(struct dquot *dquot)
270 {
271         /* We add to the back of inuse list so we don't have to restart
272          * when traversing this list and we block */
273         list_add_tail(&dquot->dq_inuse, &inuse_list);
274         dqstats.allocated_dquots++;
275 }
276
277 static inline void remove_inuse(struct dquot *dquot)
278 {
279         dqstats.allocated_dquots--;
280         list_del(&dquot->dq_inuse);
281 }
282 /*
283  * End of list functions needing dq_list_lock
284  */
285
286 static void wait_on_dquot(struct dquot *dquot)
287 {
288         mutex_lock(&dquot->dq_lock);
289         mutex_unlock(&dquot->dq_lock);
290 }
291
292 static inline int dquot_dirty(struct dquot *dquot)
293 {
294         return test_bit(DQ_MOD_B, &dquot->dq_flags);
295 }
296
297 static inline int mark_dquot_dirty(struct dquot *dquot)
298 {
299         return dquot->dq_sb->dq_op->mark_dirty(dquot);
300 }
301
302 int dquot_mark_dquot_dirty(struct dquot *dquot)
303 {
304         spin_lock(&dq_list_lock);
305         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
306                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
307                                 info[dquot->dq_type].dqi_dirty_list);
308         spin_unlock(&dq_list_lock);
309         return 0;
310 }
311
312 /* This function needs dq_list_lock */
313 static inline int clear_dquot_dirty(struct dquot *dquot)
314 {
315         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
316                 return 0;
317         list_del_init(&dquot->dq_dirty);
318         return 1;
319 }
320
321 void mark_info_dirty(struct super_block *sb, int type)
322 {
323         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
324 }
325 EXPORT_SYMBOL(mark_info_dirty);
326
327 /*
328  *      Read dquot from disk and alloc space for it
329  */
330
331 int dquot_acquire(struct dquot *dquot)
332 {
333         int ret = 0, ret2 = 0;
334         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
335
336         mutex_lock(&dquot->dq_lock);
337         mutex_lock(&dqopt->dqio_mutex);
338         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
339                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
340         if (ret < 0)
341                 goto out_iolock;
342         set_bit(DQ_READ_B, &dquot->dq_flags);
343         /* Instantiate dquot if needed */
344         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
345                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
346                 /* Write the info if needed */
347                 if (info_dirty(&dqopt->info[dquot->dq_type]))
348                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
349                 if (ret < 0)
350                         goto out_iolock;
351                 if (ret2 < 0) {
352                         ret = ret2;
353                         goto out_iolock;
354                 }
355         }
356         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
357 out_iolock:
358         mutex_unlock(&dqopt->dqio_mutex);
359         mutex_unlock(&dquot->dq_lock);
360         return ret;
361 }
362
363 /*
364  *      Write dquot to disk
365  */
366 int dquot_commit(struct dquot *dquot)
367 {
368         int ret = 0, ret2 = 0;
369         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
370
371         mutex_lock(&dqopt->dqio_mutex);
372         spin_lock(&dq_list_lock);
373         if (!clear_dquot_dirty(dquot)) {
374                 spin_unlock(&dq_list_lock);
375                 goto out_sem;
376         }
377         spin_unlock(&dq_list_lock);
378         /* Inactive dquot can be only if there was error during read/init
379          * => we have better not writing it */
380         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
381                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
382                 if (info_dirty(&dqopt->info[dquot->dq_type]))
383                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
384                 if (ret >= 0)
385                         ret = ret2;
386         }
387 out_sem:
388         mutex_unlock(&dqopt->dqio_mutex);
389         return ret;
390 }
391
392 /*
393  *      Release dquot
394  */
395 int dquot_release(struct dquot *dquot)
396 {
397         int ret = 0, ret2 = 0;
398         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
399
400         mutex_lock(&dquot->dq_lock);
401         /* Check whether we are not racing with some other dqget() */
402         if (atomic_read(&dquot->dq_count) > 1)
403                 goto out_dqlock;
404         mutex_lock(&dqopt->dqio_mutex);
405         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
406                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
407                 /* Write the info */
408                 if (info_dirty(&dqopt->info[dquot->dq_type]))
409                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
410                 if (ret >= 0)
411                         ret = ret2;
412         }
413         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
414         mutex_unlock(&dqopt->dqio_mutex);
415 out_dqlock:
416         mutex_unlock(&dquot->dq_lock);
417         return ret;
418 }
419
420 /* Invalidate all dquots on the list. Note that this function is called after
421  * quota is disabled and pointers from inodes removed so there cannot be new
422  * quota users. There can still be some users of quotas due to inodes being
423  * just deleted or pruned by prune_icache() (those are not attached to any
424  * list). We have to wait for such users.
425  */
426 static void invalidate_dquots(struct super_block *sb, int type)
427 {
428         struct dquot *dquot, *tmp;
429
430 restart:
431         spin_lock(&dq_list_lock);
432         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
433                 if (dquot->dq_sb != sb)
434                         continue;
435                 if (dquot->dq_type != type)
436                         continue;
437                 /* Wait for dquot users */
438                 if (atomic_read(&dquot->dq_count)) {
439                         DEFINE_WAIT(wait);
440
441                         atomic_inc(&dquot->dq_count);
442                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
443                                         TASK_UNINTERRUPTIBLE);
444                         spin_unlock(&dq_list_lock);
445                         /* Once dqput() wakes us up, we know it's time to free
446                          * the dquot.
447                          * IMPORTANT: we rely on the fact that there is always
448                          * at most one process waiting for dquot to free.
449                          * Otherwise dq_count would be > 1 and we would never
450                          * wake up.
451                          */
452                         if (atomic_read(&dquot->dq_count) > 1)
453                                 schedule();
454                         finish_wait(&dquot->dq_wait_unused, &wait);
455                         dqput(dquot);
456                         /* At this moment dquot() need not exist (it could be
457                          * reclaimed by prune_dqcache(). Hence we must
458                          * restart. */
459                         goto restart;
460                 }
461                 /*
462                  * Quota now has no users and it has been written on last
463                  * dqput()
464                  */
465                 remove_dquot_hash(dquot);
466                 remove_free_dquot(dquot);
467                 remove_inuse(dquot);
468                 kmem_cache_free(dquot_cachep, dquot);
469         }
470         spin_unlock(&dq_list_lock);
471 }
472
473 int vfs_quota_sync(struct super_block *sb, int type)
474 {
475         struct list_head *dirty;
476         struct dquot *dquot;
477         struct quota_info *dqopt = sb_dqopt(sb);
478         int cnt;
479
480         mutex_lock(&dqopt->dqonoff_mutex);
481         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
482                 if (type != -1 && cnt != type)
483                         continue;
484                 if (!sb_has_quota_enabled(sb, cnt))
485                         continue;
486                 spin_lock(&dq_list_lock);
487                 dirty = &dqopt->info[cnt].dqi_dirty_list;
488                 while (!list_empty(dirty)) {
489                         dquot = list_first_entry(dirty, struct dquot, dq_dirty);
490                         /* Dirty and inactive can be only bad dquot... */
491                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
492                                 clear_dquot_dirty(dquot);
493                                 continue;
494                         }
495                         /* Now we have active dquot from which someone is
496                          * holding reference so we can safely just increase
497                          * use count */
498                         atomic_inc(&dquot->dq_count);
499                         dqstats.lookups++;
500                         spin_unlock(&dq_list_lock);
501                         sb->dq_op->write_dquot(dquot);
502                         dqput(dquot);
503                         spin_lock(&dq_list_lock);
504                 }
505                 spin_unlock(&dq_list_lock);
506         }
507
508         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
509                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
510                         && info_dirty(&dqopt->info[cnt]))
511                         sb->dq_op->write_info(sb, cnt);
512         spin_lock(&dq_list_lock);
513         dqstats.syncs++;
514         spin_unlock(&dq_list_lock);
515         mutex_unlock(&dqopt->dqonoff_mutex);
516
517         return 0;
518 }
519
520 /* Free unused dquots from cache */
521 static void prune_dqcache(int count)
522 {
523         struct list_head *head;
524         struct dquot *dquot;
525
526         head = free_dquots.prev;
527         while (head != &free_dquots && count) {
528                 dquot = list_entry(head, struct dquot, dq_free);
529                 remove_dquot_hash(dquot);
530                 remove_free_dquot(dquot);
531                 remove_inuse(dquot);
532                 kmem_cache_free(dquot_cachep, dquot);
533                 count--;
534                 head = free_dquots.prev;
535         }
536 }
537
538 /*
539  * This is called from kswapd when we think we need some
540  * more memory
541  */
542
543 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
544 {
545         if (nr) {
546                 spin_lock(&dq_list_lock);
547                 prune_dqcache(nr);
548                 spin_unlock(&dq_list_lock);
549         }
550         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
551 }
552
553 static struct shrinker dqcache_shrinker = {
554         .shrink = shrink_dqcache_memory,
555         .seeks = DEFAULT_SEEKS,
556 };
557
558 /*
559  * Put reference to dquot
560  * NOTE: If you change this function please check whether dqput_blocks() works right...
561  * MUST be called with either dqptr_sem or dqonoff_mutex held
562  */
563 static void dqput(struct dquot *dquot)
564 {
565         int ret;
566
567         if (!dquot)
568                 return;
569 #ifdef __DQUOT_PARANOIA
570         if (!atomic_read(&dquot->dq_count)) {
571                 printk("VFS: dqput: trying to free free dquot\n");
572                 printk("VFS: device %s, dquot of %s %d\n",
573                         dquot->dq_sb->s_id,
574                         quotatypes[dquot->dq_type],
575                         dquot->dq_id);
576                 BUG();
577         }
578 #endif
579         
580         spin_lock(&dq_list_lock);
581         dqstats.drops++;
582         spin_unlock(&dq_list_lock);
583 we_slept:
584         spin_lock(&dq_list_lock);
585         if (atomic_read(&dquot->dq_count) > 1) {
586                 /* We have more than one user... nothing to do */
587                 atomic_dec(&dquot->dq_count);
588                 /* Releasing dquot during quotaoff phase? */
589                 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
590                     atomic_read(&dquot->dq_count) == 1)
591                         wake_up(&dquot->dq_wait_unused);
592                 spin_unlock(&dq_list_lock);
593                 return;
594         }
595         /* Need to release dquot? */
596         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
597                 spin_unlock(&dq_list_lock);
598                 /* Commit dquot before releasing */
599                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
600                 if (ret < 0) {
601                         printk(KERN_ERR "VFS: cannot write quota structure on "
602                                 "device %s (error %d). Quota may get out of "
603                                 "sync!\n", dquot->dq_sb->s_id, ret);
604                         /*
605                          * We clear dirty bit anyway, so that we avoid
606                          * infinite loop here
607                          */
608                         spin_lock(&dq_list_lock);
609                         clear_dquot_dirty(dquot);
610                         spin_unlock(&dq_list_lock);
611                 }
612                 goto we_slept;
613         }
614         /* Clear flag in case dquot was inactive (something bad happened) */
615         clear_dquot_dirty(dquot);
616         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
617                 spin_unlock(&dq_list_lock);
618                 dquot->dq_sb->dq_op->release_dquot(dquot);
619                 goto we_slept;
620         }
621         atomic_dec(&dquot->dq_count);
622 #ifdef __DQUOT_PARANOIA
623         /* sanity check */
624         BUG_ON(!list_empty(&dquot->dq_free));
625 #endif
626         put_dquot_last(dquot);
627         spin_unlock(&dq_list_lock);
628 }
629
630 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
631 {
632         struct dquot *dquot;
633
634         dquot = kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
635         if(!dquot)
636                 return NODQUOT;
637
638         mutex_init(&dquot->dq_lock);
639         INIT_LIST_HEAD(&dquot->dq_free);
640         INIT_LIST_HEAD(&dquot->dq_inuse);
641         INIT_HLIST_NODE(&dquot->dq_hash);
642         INIT_LIST_HEAD(&dquot->dq_dirty);
643         init_waitqueue_head(&dquot->dq_wait_unused);
644         dquot->dq_sb = sb;
645         dquot->dq_type = type;
646         atomic_set(&dquot->dq_count, 1);
647
648         return dquot;
649 }
650
651 /*
652  * Get reference to dquot
653  * MUST be called with either dqptr_sem or dqonoff_mutex held
654  */
655 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
656 {
657         unsigned int hashent = hashfn(sb, id, type);
658         struct dquot *dquot, *empty = NODQUOT;
659
660         if (!sb_has_quota_enabled(sb, type))
661                 return NODQUOT;
662 we_slept:
663         spin_lock(&dq_list_lock);
664         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
665                 if (empty == NODQUOT) {
666                         spin_unlock(&dq_list_lock);
667                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
668                                 schedule();     /* Try to wait for a moment... */
669                         goto we_slept;
670                 }
671                 dquot = empty;
672                 dquot->dq_id = id;
673                 /* all dquots go on the inuse_list */
674                 put_inuse(dquot);
675                 /* hash it first so it can be found */
676                 insert_dquot_hash(dquot);
677                 dqstats.lookups++;
678                 spin_unlock(&dq_list_lock);
679         } else {
680                 if (!atomic_read(&dquot->dq_count))
681                         remove_free_dquot(dquot);
682                 atomic_inc(&dquot->dq_count);
683                 dqstats.cache_hits++;
684                 dqstats.lookups++;
685                 spin_unlock(&dq_list_lock);
686                 if (empty)
687                         kmem_cache_free(dquot_cachep, empty);
688         }
689         /* Wait for dq_lock - after this we know that either dquot_release() is already
690          * finished or it will be canceled due to dq_count > 1 test */
691         wait_on_dquot(dquot);
692         /* Read the dquot and instantiate it (everything done only if needed) */
693         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
694                 dqput(dquot);
695                 return NODQUOT;
696         }
697 #ifdef __DQUOT_PARANOIA
698         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
699 #endif
700
701         return dquot;
702 }
703
704 static int dqinit_needed(struct inode *inode, int type)
705 {
706         int cnt;
707
708         if (IS_NOQUOTA(inode))
709                 return 0;
710         if (type != -1)
711                 return inode->i_dquot[type] == NODQUOT;
712         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
713                 if (inode->i_dquot[cnt] == NODQUOT)
714                         return 1;
715         return 0;
716 }
717
718 /* This routine is guarded by dqonoff_mutex mutex */
719 static void add_dquot_ref(struct super_block *sb, int type)
720 {
721         struct inode *inode, *old_inode = NULL;
722
723         spin_lock(&inode_lock);
724         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
725                 if (!atomic_read(&inode->i_writecount))
726                         continue;
727                 if (!dqinit_needed(inode, type))
728                         continue;
729                 if (inode->i_state & (I_FREEING|I_WILL_FREE))
730                         continue;
731
732                 __iget(inode);
733                 spin_unlock(&inode_lock);
734
735                 iput(old_inode);
736                 sb->dq_op->initialize(inode, type);
737                 /* We hold a reference to 'inode' so it couldn't have been
738                  * removed from s_inodes list while we dropped the inode_lock.
739                  * We cannot iput the inode now as we can be holding the last
740                  * reference and we cannot iput it under inode_lock. So we
741                  * keep the reference and iput it later. */
742                 old_inode = inode;
743                 spin_lock(&inode_lock);
744         }
745         spin_unlock(&inode_lock);
746         iput(old_inode);
747 }
748
749 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
750 static inline int dqput_blocks(struct dquot *dquot)
751 {
752         if (atomic_read(&dquot->dq_count) <= 1)
753                 return 1;
754         return 0;
755 }
756
757 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
758 /* We can't race with anybody because we hold dqptr_sem for writing... */
759 static int remove_inode_dquot_ref(struct inode *inode, int type,
760                                   struct list_head *tofree_head)
761 {
762         struct dquot *dquot = inode->i_dquot[type];
763
764         inode->i_dquot[type] = NODQUOT;
765         if (dquot != NODQUOT) {
766                 if (dqput_blocks(dquot)) {
767 #ifdef __DQUOT_PARANOIA
768                         if (atomic_read(&dquot->dq_count) != 1)
769                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
770 #endif
771                         spin_lock(&dq_list_lock);
772                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
773                         spin_unlock(&dq_list_lock);
774                         return 1;
775                 }
776                 else
777                         dqput(dquot);   /* We have guaranteed we won't block */
778         }
779         return 0;
780 }
781
782 /* Free list of dquots - called from inode.c */
783 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
784 static void put_dquot_list(struct list_head *tofree_head)
785 {
786         struct list_head *act_head;
787         struct dquot *dquot;
788
789         act_head = tofree_head->next;
790         /* So now we have dquots on the list... Just free them */
791         while (act_head != tofree_head) {
792                 dquot = list_entry(act_head, struct dquot, dq_free);
793                 act_head = act_head->next;
794                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
795                 dqput(dquot);
796         }
797 }
798
799 static void remove_dquot_ref(struct super_block *sb, int type,
800                 struct list_head *tofree_head)
801 {
802         struct inode *inode;
803
804         spin_lock(&inode_lock);
805         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
806                 if (!IS_NOQUOTA(inode))
807                         remove_inode_dquot_ref(inode, type, tofree_head);
808         }
809         spin_unlock(&inode_lock);
810 }
811
812 /* Gather all references from inodes and drop them */
813 static void drop_dquot_ref(struct super_block *sb, int type)
814 {
815         LIST_HEAD(tofree_head);
816
817         if (sb->dq_op) {
818                 down_write(&sb_dqopt(sb)->dqptr_sem);
819                 remove_dquot_ref(sb, type, &tofree_head);
820                 up_write(&sb_dqopt(sb)->dqptr_sem);
821                 put_dquot_list(&tofree_head);
822         }
823 }
824
825 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
826 {
827         dquot->dq_dqb.dqb_curinodes += number;
828 }
829
830 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
831 {
832         dquot->dq_dqb.dqb_curspace += number;
833 }
834
835 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
836 {
837         if (dquot->dq_dqb.dqb_curinodes > number)
838                 dquot->dq_dqb.dqb_curinodes -= number;
839         else
840                 dquot->dq_dqb.dqb_curinodes = 0;
841         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
842                 dquot->dq_dqb.dqb_itime = (time_t) 0;
843         clear_bit(DQ_INODES_B, &dquot->dq_flags);
844 }
845
846 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
847 {
848         if (dquot->dq_dqb.dqb_curspace > number)
849                 dquot->dq_dqb.dqb_curspace -= number;
850         else
851                 dquot->dq_dqb.dqb_curspace = 0;
852         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
853                 dquot->dq_dqb.dqb_btime = (time_t) 0;
854         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
855 }
856
857 static int warning_issued(struct dquot *dquot, const int warntype)
858 {
859         int flag = (warntype == QUOTA_NL_BHARDWARN ||
860                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
861                 ((warntype == QUOTA_NL_IHARDWARN ||
862                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
863
864         if (!flag)
865                 return 0;
866         return test_and_set_bit(flag, &dquot->dq_flags);
867 }
868
869 #ifdef CONFIG_PRINT_QUOTA_WARNING
870 static int flag_print_warnings = 1;
871
872 static inline int need_print_warning(struct dquot *dquot)
873 {
874         if (!flag_print_warnings)
875                 return 0;
876
877         switch (dquot->dq_type) {
878                 case USRQUOTA:
879                         return current->fsuid == dquot->dq_id;
880                 case GRPQUOTA:
881                         return in_group_p(dquot->dq_id);
882         }
883         return 0;
884 }
885
886 /* Print warning to user which exceeded quota */
887 static void print_warning(struct dquot *dquot, const int warntype)
888 {
889         char *msg = NULL;
890         struct tty_struct *tty;
891
892         if (!need_print_warning(dquot))
893                 return;
894
895         mutex_lock(&tty_mutex);
896         tty = get_current_tty();
897         if (!tty)
898                 goto out_lock;
899         tty_write_message(tty, dquot->dq_sb->s_id);
900         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
901                 tty_write_message(tty, ": warning, ");
902         else
903                 tty_write_message(tty, ": write failed, ");
904         tty_write_message(tty, quotatypes[dquot->dq_type]);
905         switch (warntype) {
906                 case QUOTA_NL_IHARDWARN:
907                         msg = " file limit reached.\r\n";
908                         break;
909                 case QUOTA_NL_ISOFTLONGWARN:
910                         msg = " file quota exceeded too long.\r\n";
911                         break;
912                 case QUOTA_NL_ISOFTWARN:
913                         msg = " file quota exceeded.\r\n";
914                         break;
915                 case QUOTA_NL_BHARDWARN:
916                         msg = " block limit reached.\r\n";
917                         break;
918                 case QUOTA_NL_BSOFTLONGWARN:
919                         msg = " block quota exceeded too long.\r\n";
920                         break;
921                 case QUOTA_NL_BSOFTWARN:
922                         msg = " block quota exceeded.\r\n";
923                         break;
924         }
925         tty_write_message(tty, msg);
926 out_lock:
927         mutex_unlock(&tty_mutex);
928 }
929 #endif
930
931 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
932
933 /* Netlink family structure for quota */
934 static struct genl_family quota_genl_family = {
935         .id = GENL_ID_GENERATE,
936         .hdrsize = 0,
937         .name = "VFS_DQUOT",
938         .version = 1,
939         .maxattr = QUOTA_NL_A_MAX,
940 };
941
942 /* Send warning to userspace about user which exceeded quota */
943 static void send_warning(const struct dquot *dquot, const char warntype)
944 {
945         static atomic_t seq;
946         struct sk_buff *skb;
947         void *msg_head;
948         int ret;
949         int msg_size = 4 * nla_total_size(sizeof(u32)) +
950                        2 * nla_total_size(sizeof(u64));
951
952         /* We have to allocate using GFP_NOFS as we are called from a
953          * filesystem performing write and thus further recursion into
954          * the fs to free some data could cause deadlocks. */
955         skb = genlmsg_new(msg_size, GFP_NOFS);
956         if (!skb) {
957                 printk(KERN_ERR
958                   "VFS: Not enough memory to send quota warning.\n");
959                 return;
960         }
961         msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
962                         &quota_genl_family, 0, QUOTA_NL_C_WARNING);
963         if (!msg_head) {
964                 printk(KERN_ERR
965                   "VFS: Cannot store netlink header in quota warning.\n");
966                 goto err_out;
967         }
968         ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
969         if (ret)
970                 goto attr_err_out;
971         ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
972         if (ret)
973                 goto attr_err_out;
974         ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
975         if (ret)
976                 goto attr_err_out;
977         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
978                 MAJOR(dquot->dq_sb->s_dev));
979         if (ret)
980                 goto attr_err_out;
981         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
982                 MINOR(dquot->dq_sb->s_dev));
983         if (ret)
984                 goto attr_err_out;
985         ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current->user->uid);
986         if (ret)
987                 goto attr_err_out;
988         genlmsg_end(skb, msg_head);
989
990         ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
991         if (ret < 0 && ret != -ESRCH)
992                 printk(KERN_ERR
993                         "VFS: Failed to send notification message: %d\n", ret);
994         return;
995 attr_err_out:
996         printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
997 err_out:
998         kfree_skb(skb);
999 }
1000 #endif
1001
1002 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
1003 {
1004         int i;
1005
1006         for (i = 0; i < MAXQUOTAS; i++)
1007                 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
1008                     !warning_issued(dquots[i], warntype[i])) {
1009 #ifdef CONFIG_PRINT_QUOTA_WARNING
1010                         print_warning(dquots[i], warntype[i]);
1011 #endif
1012 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1013                         send_warning(dquots[i], warntype[i]);
1014 #endif
1015                 }
1016 }
1017
1018 static inline char ignore_hardlimit(struct dquot *dquot)
1019 {
1020         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1021
1022         return capable(CAP_SYS_RESOURCE) &&
1023             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1024 }
1025
1026 /* needs dq_data_lock */
1027 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
1028 {
1029         *warntype = QUOTA_NL_NOWARN;
1030         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1031                 return QUOTA_OK;
1032
1033         if (dquot->dq_dqb.dqb_ihardlimit &&
1034            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1035             !ignore_hardlimit(dquot)) {
1036                 *warntype = QUOTA_NL_IHARDWARN;
1037                 return NO_QUOTA;
1038         }
1039
1040         if (dquot->dq_dqb.dqb_isoftlimit &&
1041            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1042             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1043             !ignore_hardlimit(dquot)) {
1044                 *warntype = QUOTA_NL_ISOFTLONGWARN;
1045                 return NO_QUOTA;
1046         }
1047
1048         if (dquot->dq_dqb.dqb_isoftlimit &&
1049            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1050             dquot->dq_dqb.dqb_itime == 0) {
1051                 *warntype = QUOTA_NL_ISOFTWARN;
1052                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1053         }
1054
1055         return QUOTA_OK;
1056 }
1057
1058 /* needs dq_data_lock */
1059 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1060 {
1061         *warntype = QUOTA_NL_NOWARN;
1062         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1063                 return QUOTA_OK;
1064
1065         if (dquot->dq_dqb.dqb_bhardlimit &&
1066            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
1067             !ignore_hardlimit(dquot)) {
1068                 if (!prealloc)
1069                         *warntype = QUOTA_NL_BHARDWARN;
1070                 return NO_QUOTA;
1071         }
1072
1073         if (dquot->dq_dqb.dqb_bsoftlimit &&
1074            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1075             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1076             !ignore_hardlimit(dquot)) {
1077                 if (!prealloc)
1078                         *warntype = QUOTA_NL_BSOFTLONGWARN;
1079                 return NO_QUOTA;
1080         }
1081
1082         if (dquot->dq_dqb.dqb_bsoftlimit &&
1083            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1084             dquot->dq_dqb.dqb_btime == 0) {
1085                 if (!prealloc) {
1086                         *warntype = QUOTA_NL_BSOFTWARN;
1087                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1088                 }
1089                 else
1090                         /*
1091                          * We don't allow preallocation to exceed softlimit so exceeding will
1092                          * be always printed
1093                          */
1094                         return NO_QUOTA;
1095         }
1096
1097         return QUOTA_OK;
1098 }
1099
1100 /*
1101  *      Initialize quota pointers in inode
1102  *      Transaction must be started at entry
1103  */
1104 int dquot_initialize(struct inode *inode, int type)
1105 {
1106         unsigned int id = 0;
1107         int cnt, ret = 0;
1108
1109         /* First test before acquiring mutex - solves deadlocks when we
1110          * re-enter the quota code and are already holding the mutex */
1111         if (IS_NOQUOTA(inode))
1112                 return 0;
1113         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1114         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1115         if (IS_NOQUOTA(inode))
1116                 goto out_err;
1117         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1118                 if (type != -1 && cnt != type)
1119                         continue;
1120                 if (inode->i_dquot[cnt] == NODQUOT) {
1121                         switch (cnt) {
1122                                 case USRQUOTA:
1123                                         id = inode->i_uid;
1124                                         break;
1125                                 case GRPQUOTA:
1126                                         id = inode->i_gid;
1127                                         break;
1128                         }
1129                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1130                 }
1131         }
1132 out_err:
1133         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1134         return ret;
1135 }
1136
1137 /*
1138  *      Release all quotas referenced by inode
1139  *      Transaction must be started at an entry
1140  */
1141 int dquot_drop(struct inode *inode)
1142 {
1143         int cnt;
1144
1145         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1146         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1147                 if (inode->i_dquot[cnt] != NODQUOT) {
1148                         dqput(inode->i_dquot[cnt]);
1149                         inode->i_dquot[cnt] = NODQUOT;
1150                 }
1151         }
1152         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1153         return 0;
1154 }
1155
1156 /* Wrapper to remove references to quota structures from inode */
1157 void vfs_dq_drop(struct inode *inode)
1158 {
1159         /* Here we can get arbitrary inode from clear_inode() so we have
1160          * to be careful. OTOH we don't need locking as quota operations
1161          * are allowed to change only at mount time */
1162         if (!IS_NOQUOTA(inode) && inode->i_sb && inode->i_sb->dq_op
1163             && inode->i_sb->dq_op->drop) {
1164                 int cnt;
1165                 /* Test before calling to rule out calls from proc and such
1166                  * where we are not allowed to block. Note that this is
1167                  * actually reliable test even without the lock - the caller
1168                  * must assure that nobody can come after the DQUOT_DROP and
1169                  * add quota pointers back anyway */
1170                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1171                         if (inode->i_dquot[cnt] != NODQUOT)
1172                                 break;
1173                 if (cnt < MAXQUOTAS)
1174                         inode->i_sb->dq_op->drop(inode);
1175         }
1176 }
1177
1178 /*
1179  * Following four functions update i_blocks+i_bytes fields and
1180  * quota information (together with appropriate checks)
1181  * NOTE: We absolutely rely on the fact that caller dirties
1182  * the inode (usually macros in quotaops.h care about this) and
1183  * holds a handle for the current transaction so that dquot write and
1184  * inode write go into the same transaction.
1185  */
1186
1187 /*
1188  * This operation can block, but only after everything is updated
1189  */
1190 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1191 {
1192         int cnt, ret = NO_QUOTA;
1193         char warntype[MAXQUOTAS];
1194
1195         /* First test before acquiring mutex - solves deadlocks when we
1196          * re-enter the quota code and are already holding the mutex */
1197         if (IS_NOQUOTA(inode)) {
1198 out_add:
1199                 inode_add_bytes(inode, number);
1200                 return QUOTA_OK;
1201         }
1202         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1203                 warntype[cnt] = QUOTA_NL_NOWARN;
1204
1205         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1206         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1207                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1208                 goto out_add;
1209         }
1210         spin_lock(&dq_data_lock);
1211         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1212                 if (inode->i_dquot[cnt] == NODQUOT)
1213                         continue;
1214                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1215                         goto warn_put_all;
1216         }
1217         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1218                 if (inode->i_dquot[cnt] == NODQUOT)
1219                         continue;
1220                 dquot_incr_space(inode->i_dquot[cnt], number);
1221         }
1222         inode_add_bytes(inode, number);
1223         ret = QUOTA_OK;
1224 warn_put_all:
1225         spin_unlock(&dq_data_lock);
1226         if (ret == QUOTA_OK)
1227                 /* Dirtify all the dquots - this can block when journalling */
1228                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1229                         if (inode->i_dquot[cnt])
1230                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1231         flush_warnings(inode->i_dquot, warntype);
1232         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1233         return ret;
1234 }
1235
1236 /*
1237  * This operation can block, but only after everything is updated
1238  */
1239 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1240 {
1241         int cnt, ret = NO_QUOTA;
1242         char warntype[MAXQUOTAS];
1243
1244         /* First test before acquiring mutex - solves deadlocks when we
1245          * re-enter the quota code and are already holding the mutex */
1246         if (IS_NOQUOTA(inode))
1247                 return QUOTA_OK;
1248         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1249                 warntype[cnt] = QUOTA_NL_NOWARN;
1250         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1251         if (IS_NOQUOTA(inode)) {
1252                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1253                 return QUOTA_OK;
1254         }
1255         spin_lock(&dq_data_lock);
1256         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1257                 if (inode->i_dquot[cnt] == NODQUOT)
1258                         continue;
1259                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1260                         goto warn_put_all;
1261         }
1262
1263         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1264                 if (inode->i_dquot[cnt] == NODQUOT)
1265                         continue;
1266                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1267         }
1268         ret = QUOTA_OK;
1269 warn_put_all:
1270         spin_unlock(&dq_data_lock);
1271         if (ret == QUOTA_OK)
1272                 /* Dirtify all the dquots - this can block when journalling */
1273                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1274                         if (inode->i_dquot[cnt])
1275                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1276         flush_warnings(inode->i_dquot, warntype);
1277         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1278         return ret;
1279 }
1280
1281 /*
1282  * This operation can block, but only after everything is updated
1283  */
1284 int dquot_free_space(struct inode *inode, qsize_t number)
1285 {
1286         unsigned int cnt;
1287
1288         /* First test before acquiring mutex - solves deadlocks when we
1289          * re-enter the quota code and are already holding the mutex */
1290         if (IS_NOQUOTA(inode)) {
1291 out_sub:
1292                 inode_sub_bytes(inode, number);
1293                 return QUOTA_OK;
1294         }
1295         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1296         /* Now recheck reliably when holding dqptr_sem */
1297         if (IS_NOQUOTA(inode)) {
1298                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1299                 goto out_sub;
1300         }
1301         spin_lock(&dq_data_lock);
1302         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1303                 if (inode->i_dquot[cnt] == NODQUOT)
1304                         continue;
1305                 dquot_decr_space(inode->i_dquot[cnt], number);
1306         }
1307         inode_sub_bytes(inode, number);
1308         spin_unlock(&dq_data_lock);
1309         /* Dirtify all the dquots - this can block when journalling */
1310         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1311                 if (inode->i_dquot[cnt])
1312                         mark_dquot_dirty(inode->i_dquot[cnt]);
1313         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1314         return QUOTA_OK;
1315 }
1316
1317 /*
1318  * This operation can block, but only after everything is updated
1319  */
1320 int dquot_free_inode(const struct inode *inode, unsigned long number)
1321 {
1322         unsigned int cnt;
1323
1324         /* First test before acquiring mutex - solves deadlocks when we
1325          * re-enter the quota code and are already holding the mutex */
1326         if (IS_NOQUOTA(inode))
1327                 return QUOTA_OK;
1328         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1329         /* Now recheck reliably when holding dqptr_sem */
1330         if (IS_NOQUOTA(inode)) {
1331                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1332                 return QUOTA_OK;
1333         }
1334         spin_lock(&dq_data_lock);
1335         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1336                 if (inode->i_dquot[cnt] == NODQUOT)
1337                         continue;
1338                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1339         }
1340         spin_unlock(&dq_data_lock);
1341         /* Dirtify all the dquots - this can block when journalling */
1342         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1343                 if (inode->i_dquot[cnt])
1344                         mark_dquot_dirty(inode->i_dquot[cnt]);
1345         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1346         return QUOTA_OK;
1347 }
1348
1349 /*
1350  * Transfer the number of inode and blocks from one diskquota to an other.
1351  *
1352  * This operation can block, but only after everything is updated
1353  * A transaction must be started when entering this function.
1354  */
1355 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1356 {
1357         qsize_t space;
1358         struct dquot *transfer_from[MAXQUOTAS];
1359         struct dquot *transfer_to[MAXQUOTAS];
1360         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1361             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1362         char warntype[MAXQUOTAS];
1363
1364         /* First test before acquiring mutex - solves deadlocks when we
1365          * re-enter the quota code and are already holding the mutex */
1366         if (IS_NOQUOTA(inode))
1367                 return QUOTA_OK;
1368         /* Clear the arrays */
1369         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1370                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1371                 warntype[cnt] = QUOTA_NL_NOWARN;
1372         }
1373         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1374         /* Now recheck reliably when holding dqptr_sem */
1375         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1376                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1377                 return QUOTA_OK;
1378         }
1379         /* First build the transfer_to list - here we can block on
1380          * reading/instantiating of dquots.  We know that the transaction for
1381          * us was already started so we don't violate lock ranking here */
1382         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1383                 switch (cnt) {
1384                         case USRQUOTA:
1385                                 if (!chuid)
1386                                         continue;
1387                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1388                                 break;
1389                         case GRPQUOTA:
1390                                 if (!chgid)
1391                                         continue;
1392                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1393                                 break;
1394                 }
1395         }
1396         spin_lock(&dq_data_lock);
1397         space = inode_get_bytes(inode);
1398         /* Build the transfer_from list and check the limits */
1399         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1400                 if (transfer_to[cnt] == NODQUOT)
1401                         continue;
1402                 transfer_from[cnt] = inode->i_dquot[cnt];
1403                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1404                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1405                         goto warn_put_all;
1406         }
1407
1408         /*
1409          * Finally perform the needed transfer from transfer_from to transfer_to
1410          */
1411         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1412                 /*
1413                  * Skip changes for same uid or gid or for turned off quota-type.
1414                  */
1415                 if (transfer_to[cnt] == NODQUOT)
1416                         continue;
1417
1418                 /* Due to IO error we might not have transfer_from[] structure */
1419                 if (transfer_from[cnt]) {
1420                         dquot_decr_inodes(transfer_from[cnt], 1);
1421                         dquot_decr_space(transfer_from[cnt], space);
1422                 }
1423
1424                 dquot_incr_inodes(transfer_to[cnt], 1);
1425                 dquot_incr_space(transfer_to[cnt], space);
1426
1427                 inode->i_dquot[cnt] = transfer_to[cnt];
1428         }
1429         ret = QUOTA_OK;
1430 warn_put_all:
1431         spin_unlock(&dq_data_lock);
1432         /* Dirtify all the dquots - this can block when journalling */
1433         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1434                 if (transfer_from[cnt])
1435                         mark_dquot_dirty(transfer_from[cnt]);
1436                 if (transfer_to[cnt])
1437                         mark_dquot_dirty(transfer_to[cnt]);
1438         }
1439         flush_warnings(transfer_to, warntype);
1440         
1441         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1442                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1443                         dqput(transfer_from[cnt]);
1444                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1445                         dqput(transfer_to[cnt]);
1446         }
1447         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1448         return ret;
1449 }
1450
1451 /* Wrapper for transferring ownership of an inode */
1452 int vfs_dq_transfer(struct inode *inode, struct iattr *iattr)
1453 {
1454         if (sb_any_quota_enabled(inode->i_sb) && !IS_NOQUOTA(inode)) {
1455                 vfs_dq_init(inode);
1456                 if (inode->i_sb->dq_op->transfer(inode, iattr) == NO_QUOTA)
1457                         return 1;
1458         }
1459         return 0;
1460 }
1461
1462
1463 /*
1464  * Write info of quota file to disk
1465  */
1466 int dquot_commit_info(struct super_block *sb, int type)
1467 {
1468         int ret;
1469         struct quota_info *dqopt = sb_dqopt(sb);
1470
1471         mutex_lock(&dqopt->dqio_mutex);
1472         ret = dqopt->ops[type]->write_file_info(sb, type);
1473         mutex_unlock(&dqopt->dqio_mutex);
1474         return ret;
1475 }
1476
1477 /*
1478  * Definitions of diskquota operations.
1479  */
1480 struct dquot_operations dquot_operations = {
1481         .initialize     = dquot_initialize,
1482         .drop           = dquot_drop,
1483         .alloc_space    = dquot_alloc_space,
1484         .alloc_inode    = dquot_alloc_inode,
1485         .free_space     = dquot_free_space,
1486         .free_inode     = dquot_free_inode,
1487         .transfer       = dquot_transfer,
1488         .write_dquot    = dquot_commit,
1489         .acquire_dquot  = dquot_acquire,
1490         .release_dquot  = dquot_release,
1491         .mark_dirty     = dquot_mark_dquot_dirty,
1492         .write_info     = dquot_commit_info
1493 };
1494
1495 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1496 {
1497         switch (type) {
1498                 case USRQUOTA:
1499                         dqopt->flags |= DQUOT_USR_ENABLED;
1500                         dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1501                         break;
1502                 case GRPQUOTA:
1503                         dqopt->flags |= DQUOT_GRP_ENABLED;
1504                         dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1505                         break;
1506         }
1507 }
1508
1509 static inline void reset_enable_flags(struct quota_info *dqopt, int type,
1510                                       int remount)
1511 {
1512         switch (type) {
1513                 case USRQUOTA:
1514                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1515                         if (remount)
1516                                 dqopt->flags |= DQUOT_USR_SUSPENDED;
1517                         else
1518                                 dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1519                         break;
1520                 case GRPQUOTA:
1521                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1522                         if (remount)
1523                                 dqopt->flags |= DQUOT_GRP_SUSPENDED;
1524                         else
1525                                 dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1526                         break;
1527         }
1528 }
1529
1530
1531 /*
1532  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1533  */
1534 int vfs_quota_off(struct super_block *sb, int type, int remount)
1535 {
1536         int cnt, ret = 0;
1537         struct quota_info *dqopt = sb_dqopt(sb);
1538         struct inode *toputinode[MAXQUOTAS];
1539
1540         /* We need to serialize quota_off() for device */
1541         mutex_lock(&dqopt->dqonoff_mutex);
1542
1543         /*
1544          * Skip everything if there's nothing to do. We have to do this because
1545          * sometimes we are called when fill_super() failed and calling
1546          * sync_fs() in such cases does no good.
1547          */
1548         if (!sb_any_quota_enabled(sb) && !sb_any_quota_suspended(sb)) {
1549                 mutex_unlock(&dqopt->dqonoff_mutex);
1550                 return 0;
1551         }
1552         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1553                 toputinode[cnt] = NULL;
1554                 if (type != -1 && cnt != type)
1555                         continue;
1556                 /* If we keep inodes of quota files after remount and quotaoff
1557                  * is called, drop kept inodes. */
1558                 if (!remount && sb_has_quota_suspended(sb, cnt)) {
1559                         iput(dqopt->files[cnt]);
1560                         dqopt->files[cnt] = NULL;
1561                         reset_enable_flags(dqopt, cnt, 0);
1562                         continue;
1563                 }
1564                 if (!sb_has_quota_enabled(sb, cnt))
1565                         continue;
1566                 reset_enable_flags(dqopt, cnt, remount);
1567
1568                 /* Note: these are blocking operations */
1569                 drop_dquot_ref(sb, cnt);
1570                 invalidate_dquots(sb, cnt);
1571                 /*
1572                  * Now all dquots should be invalidated, all writes done so we should be only
1573                  * users of the info. No locks needed.
1574                  */
1575                 if (info_dirty(&dqopt->info[cnt]))
1576                         sb->dq_op->write_info(sb, cnt);
1577                 if (dqopt->ops[cnt]->free_file_info)
1578                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1579                 put_quota_format(dqopt->info[cnt].dqi_format);
1580
1581                 toputinode[cnt] = dqopt->files[cnt];
1582                 if (!remount)
1583                         dqopt->files[cnt] = NULL;
1584                 dqopt->info[cnt].dqi_flags = 0;
1585                 dqopt->info[cnt].dqi_igrace = 0;
1586                 dqopt->info[cnt].dqi_bgrace = 0;
1587                 dqopt->ops[cnt] = NULL;
1588         }
1589         mutex_unlock(&dqopt->dqonoff_mutex);
1590         /* Sync the superblock so that buffers with quota data are written to
1591          * disk (and so userspace sees correct data afterwards). */
1592         if (sb->s_op->sync_fs)
1593                 sb->s_op->sync_fs(sb, 1);
1594         sync_blockdev(sb->s_bdev);
1595         /* Now the quota files are just ordinary files and we can set the
1596          * inode flags back. Moreover we discard the pagecache so that
1597          * userspace sees the writes we did bypassing the pagecache. We
1598          * must also discard the blockdev buffers so that we see the
1599          * changes done by userspace on the next quotaon() */
1600         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1601                 if (toputinode[cnt]) {
1602                         mutex_lock(&dqopt->dqonoff_mutex);
1603                         /* If quota was reenabled in the meantime, we have
1604                          * nothing to do */
1605                         if (!sb_has_quota_enabled(sb, cnt)) {
1606                                 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1607                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1608                                   S_NOATIME | S_NOQUOTA);
1609                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1610                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1611                                 mark_inode_dirty(toputinode[cnt]);
1612                         }
1613                         mutex_unlock(&dqopt->dqonoff_mutex);
1614                         /* On remount RO, we keep the inode pointer so that we
1615                          * can reenable quota on the subsequent remount RW.
1616                          * But we have better not keep inode pointer when there
1617                          * is pending delete on the quota file... */
1618                         if (!remount)
1619                                 iput(toputinode[cnt]);
1620                         else if (!toputinode[cnt]->i_nlink)
1621                                 ret = -EBUSY;
1622                 }
1623         if (sb->s_bdev)
1624                 invalidate_bdev(sb->s_bdev);
1625         return ret;
1626 }
1627
1628 /*
1629  *      Turn quotas on on a device
1630  */
1631
1632 /* Helper function when we already have the inode */
1633 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1634 {
1635         struct quota_format_type *fmt = find_quota_format(format_id);
1636         struct super_block *sb = inode->i_sb;
1637         struct quota_info *dqopt = sb_dqopt(sb);
1638         int error;
1639         int oldflags = -1;
1640
1641         if (!fmt)
1642                 return -ESRCH;
1643         if (!S_ISREG(inode->i_mode)) {
1644                 error = -EACCES;
1645                 goto out_fmt;
1646         }
1647         if (IS_RDONLY(inode)) {
1648                 error = -EROFS;
1649                 goto out_fmt;
1650         }
1651         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1652                 error = -EINVAL;
1653                 goto out_fmt;
1654         }
1655
1656         /* As we bypass the pagecache we must now flush the inode so that
1657          * we see all the changes from userspace... */
1658         write_inode_now(inode, 1);
1659         /* And now flush the block cache so that kernel sees the changes */
1660         invalidate_bdev(sb->s_bdev);
1661         mutex_lock(&inode->i_mutex);
1662         mutex_lock(&dqopt->dqonoff_mutex);
1663         if (sb_has_quota_enabled(sb, type) ||
1664                         sb_has_quota_suspended(sb, type)) {
1665                 error = -EBUSY;
1666                 goto out_lock;
1667         }
1668         /* We don't want quota and atime on quota files (deadlocks possible)
1669          * Also nobody should write to the file - we use special IO operations
1670          * which ignore the immutable bit. */
1671         down_write(&dqopt->dqptr_sem);
1672         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1673         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1674         up_write(&dqopt->dqptr_sem);
1675         sb->dq_op->drop(inode);
1676
1677         error = -EIO;
1678         dqopt->files[type] = igrab(inode);
1679         if (!dqopt->files[type])
1680                 goto out_lock;
1681         error = -EINVAL;
1682         if (!fmt->qf_ops->check_quota_file(sb, type))
1683                 goto out_file_init;
1684
1685         dqopt->ops[type] = fmt->qf_ops;
1686         dqopt->info[type].dqi_format = fmt;
1687         dqopt->info[type].dqi_fmt_id = format_id;
1688         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1689         mutex_lock(&dqopt->dqio_mutex);
1690         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1691                 mutex_unlock(&dqopt->dqio_mutex);
1692                 goto out_file_init;
1693         }
1694         mutex_unlock(&dqopt->dqio_mutex);
1695         mutex_unlock(&inode->i_mutex);
1696         set_enable_flags(dqopt, type);
1697
1698         add_dquot_ref(sb, type);
1699         mutex_unlock(&dqopt->dqonoff_mutex);
1700
1701         return 0;
1702
1703 out_file_init:
1704         dqopt->files[type] = NULL;
1705         iput(inode);
1706 out_lock:
1707         mutex_unlock(&dqopt->dqonoff_mutex);
1708         if (oldflags != -1) {
1709                 down_write(&dqopt->dqptr_sem);
1710                 /* Set the flags back (in the case of accidental quotaon()
1711                  * on a wrong file we don't want to mess up the flags) */
1712                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1713                 inode->i_flags |= oldflags;
1714                 up_write(&dqopt->dqptr_sem);
1715         }
1716         mutex_unlock(&inode->i_mutex);
1717 out_fmt:
1718         put_quota_format(fmt);
1719
1720         return error; 
1721 }
1722
1723 /* Reenable quotas on remount RW */
1724 static int vfs_quota_on_remount(struct super_block *sb, int type)
1725 {
1726         struct quota_info *dqopt = sb_dqopt(sb);
1727         struct inode *inode;
1728         int ret;
1729
1730         mutex_lock(&dqopt->dqonoff_mutex);
1731         if (!sb_has_quota_suspended(sb, type)) {
1732                 mutex_unlock(&dqopt->dqonoff_mutex);
1733                 return 0;
1734         }
1735         BUG_ON(sb_has_quota_enabled(sb, type));
1736
1737         inode = dqopt->files[type];
1738         dqopt->files[type] = NULL;
1739         reset_enable_flags(dqopt, type, 0);
1740         mutex_unlock(&dqopt->dqonoff_mutex);
1741
1742         ret = vfs_quota_on_inode(inode, type, dqopt->info[type].dqi_fmt_id);
1743         iput(inode);
1744
1745         return ret;
1746 }
1747
1748 /* Actual function called from quotactl() */
1749 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path,
1750                  int remount)
1751 {
1752         struct nameidata nd;
1753         int error;
1754
1755         if (remount)
1756                 return vfs_quota_on_remount(sb, type);
1757
1758         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1759         if (error < 0)
1760                 return error;
1761         error = security_quota_on(nd.path.dentry);
1762         if (error)
1763                 goto out_path;
1764         /* Quota file not on the same filesystem? */
1765         if (nd.path.mnt->mnt_sb != sb)
1766                 error = -EXDEV;
1767         else
1768                 error = vfs_quota_on_inode(nd.path.dentry->d_inode, type,
1769                                            format_id);
1770 out_path:
1771         path_put(&nd.path);
1772         return error;
1773 }
1774
1775 /*
1776  * This function is used when filesystem needs to initialize quotas
1777  * during mount time.
1778  */
1779 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1780                 int format_id, int type)
1781 {
1782         struct dentry *dentry;
1783         int error;
1784
1785         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1786         if (IS_ERR(dentry))
1787                 return PTR_ERR(dentry);
1788
1789         if (!dentry->d_inode) {
1790                 error = -ENOENT;
1791                 goto out;
1792         }
1793
1794         error = security_quota_on(dentry);
1795         if (!error)
1796                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1797
1798 out:
1799         dput(dentry);
1800         return error;
1801 }
1802
1803 /* Wrapper to turn on quotas when remounting rw */
1804 int vfs_dq_quota_on_remount(struct super_block *sb)
1805 {
1806         int cnt;
1807         int ret = 0, err;
1808
1809         if (!sb->s_qcop || !sb->s_qcop->quota_on)
1810                 return -ENOSYS;
1811         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1812                 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
1813                 if (err < 0 && !ret)
1814                         ret = err;
1815         }
1816         return ret;
1817 }
1818
1819 /* Generic routine for getting common part of quota structure */
1820 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1821 {
1822         struct mem_dqblk *dm = &dquot->dq_dqb;
1823
1824         spin_lock(&dq_data_lock);
1825         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1826         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1827         di->dqb_curspace = dm->dqb_curspace;
1828         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1829         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1830         di->dqb_curinodes = dm->dqb_curinodes;
1831         di->dqb_btime = dm->dqb_btime;
1832         di->dqb_itime = dm->dqb_itime;
1833         di->dqb_valid = QIF_ALL;
1834         spin_unlock(&dq_data_lock);
1835 }
1836
1837 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1838 {
1839         struct dquot *dquot;
1840
1841         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1842         if (!(dquot = dqget(sb, id, type))) {
1843                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1844                 return -ESRCH;
1845         }
1846         do_get_dqblk(dquot, di);
1847         dqput(dquot);
1848         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1849         return 0;
1850 }
1851
1852 /* Generic routine for setting common part of quota structure */
1853 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1854 {
1855         struct mem_dqblk *dm = &dquot->dq_dqb;
1856         int check_blim = 0, check_ilim = 0;
1857         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1858
1859         if ((di->dqb_valid & QIF_BLIMITS &&
1860              (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
1861               di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
1862             (di->dqb_valid & QIF_ILIMITS &&
1863              (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
1864               di->dqb_isoftlimit > dqi->dqi_maxilimit)))
1865                 return -ERANGE;
1866
1867         spin_lock(&dq_data_lock);
1868         if (di->dqb_valid & QIF_SPACE) {
1869                 dm->dqb_curspace = di->dqb_curspace;
1870                 check_blim = 1;
1871         }
1872         if (di->dqb_valid & QIF_BLIMITS) {
1873                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1874                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1875                 check_blim = 1;
1876         }
1877         if (di->dqb_valid & QIF_INODES) {
1878                 dm->dqb_curinodes = di->dqb_curinodes;
1879                 check_ilim = 1;
1880         }
1881         if (di->dqb_valid & QIF_ILIMITS) {
1882                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1883                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1884                 check_ilim = 1;
1885         }
1886         if (di->dqb_valid & QIF_BTIME)
1887                 dm->dqb_btime = di->dqb_btime;
1888         if (di->dqb_valid & QIF_ITIME)
1889                 dm->dqb_itime = di->dqb_itime;
1890
1891         if (check_blim) {
1892                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1893                         dm->dqb_btime = 0;
1894                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1895                 }
1896                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1897                         dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
1898         }
1899         if (check_ilim) {
1900                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1901                         dm->dqb_itime = 0;
1902                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1903                 }
1904                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1905                         dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
1906         }
1907         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1908                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1909         else
1910                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1911         spin_unlock(&dq_data_lock);
1912         mark_dquot_dirty(dquot);
1913
1914         return 0;
1915 }
1916
1917 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1918 {
1919         struct dquot *dquot;
1920         int rc;
1921
1922         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1923         if (!(dquot = dqget(sb, id, type))) {
1924                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1925                 return -ESRCH;
1926         }
1927         rc = do_set_dqblk(dquot, di);
1928         dqput(dquot);
1929         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1930         return rc;
1931 }
1932
1933 /* Generic routine for getting common part of quota file information */
1934 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1935 {
1936         struct mem_dqinfo *mi;
1937   
1938         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1939         if (!sb_has_quota_enabled(sb, type)) {
1940                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1941                 return -ESRCH;
1942         }
1943         mi = sb_dqopt(sb)->info + type;
1944         spin_lock(&dq_data_lock);
1945         ii->dqi_bgrace = mi->dqi_bgrace;
1946         ii->dqi_igrace = mi->dqi_igrace;
1947         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1948         ii->dqi_valid = IIF_ALL;
1949         spin_unlock(&dq_data_lock);
1950         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1951         return 0;
1952 }
1953
1954 /* Generic routine for setting common part of quota file information */
1955 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1956 {
1957         struct mem_dqinfo *mi;
1958
1959         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1960         if (!sb_has_quota_enabled(sb, type)) {
1961                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1962                 return -ESRCH;
1963         }
1964         mi = sb_dqopt(sb)->info + type;
1965         spin_lock(&dq_data_lock);
1966         if (ii->dqi_valid & IIF_BGRACE)
1967                 mi->dqi_bgrace = ii->dqi_bgrace;
1968         if (ii->dqi_valid & IIF_IGRACE)
1969                 mi->dqi_igrace = ii->dqi_igrace;
1970         if (ii->dqi_valid & IIF_FLAGS)
1971                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1972         spin_unlock(&dq_data_lock);
1973         mark_info_dirty(sb, type);
1974         /* Force write to disk */
1975         sb->dq_op->write_info(sb, type);
1976         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1977         return 0;
1978 }
1979
1980 struct quotactl_ops vfs_quotactl_ops = {
1981         .quota_on       = vfs_quota_on,
1982         .quota_off      = vfs_quota_off,
1983         .quota_sync     = vfs_quota_sync,
1984         .get_info       = vfs_get_dqinfo,
1985         .set_info       = vfs_set_dqinfo,
1986         .get_dqblk      = vfs_get_dqblk,
1987         .set_dqblk      = vfs_set_dqblk
1988 };
1989
1990 static ctl_table fs_dqstats_table[] = {
1991         {
1992                 .ctl_name       = FS_DQ_LOOKUPS,
1993                 .procname       = "lookups",
1994                 .data           = &dqstats.lookups,
1995                 .maxlen         = sizeof(int),
1996                 .mode           = 0444,
1997                 .proc_handler   = &proc_dointvec,
1998         },
1999         {
2000                 .ctl_name       = FS_DQ_DROPS,
2001                 .procname       = "drops",
2002                 .data           = &dqstats.drops,
2003                 .maxlen         = sizeof(int),
2004                 .mode           = 0444,
2005                 .proc_handler   = &proc_dointvec,
2006         },
2007         {
2008                 .ctl_name       = FS_DQ_READS,
2009                 .procname       = "reads",
2010                 .data           = &dqstats.reads,
2011                 .maxlen         = sizeof(int),
2012                 .mode           = 0444,
2013                 .proc_handler   = &proc_dointvec,
2014         },
2015         {
2016                 .ctl_name       = FS_DQ_WRITES,
2017                 .procname       = "writes",
2018                 .data           = &dqstats.writes,
2019                 .maxlen         = sizeof(int),
2020                 .mode           = 0444,
2021                 .proc_handler   = &proc_dointvec,
2022         },
2023         {
2024                 .ctl_name       = FS_DQ_CACHE_HITS,
2025                 .procname       = "cache_hits",
2026                 .data           = &dqstats.cache_hits,
2027                 .maxlen         = sizeof(int),
2028                 .mode           = 0444,
2029                 .proc_handler   = &proc_dointvec,
2030         },
2031         {
2032                 .ctl_name       = FS_DQ_ALLOCATED,
2033                 .procname       = "allocated_dquots",
2034                 .data           = &dqstats.allocated_dquots,
2035                 .maxlen         = sizeof(int),
2036                 .mode           = 0444,
2037                 .proc_handler   = &proc_dointvec,
2038         },
2039         {
2040                 .ctl_name       = FS_DQ_FREE,
2041                 .procname       = "free_dquots",
2042                 .data           = &dqstats.free_dquots,
2043                 .maxlen         = sizeof(int),
2044                 .mode           = 0444,
2045                 .proc_handler   = &proc_dointvec,
2046         },
2047         {
2048                 .ctl_name       = FS_DQ_SYNCS,
2049                 .procname       = "syncs",
2050                 .data           = &dqstats.syncs,
2051                 .maxlen         = sizeof(int),
2052                 .mode           = 0444,
2053                 .proc_handler   = &proc_dointvec,
2054         },
2055 #ifdef CONFIG_PRINT_QUOTA_WARNING
2056         {
2057                 .ctl_name       = FS_DQ_WARNINGS,
2058                 .procname       = "warnings",
2059                 .data           = &flag_print_warnings,
2060                 .maxlen         = sizeof(int),
2061                 .mode           = 0644,
2062                 .proc_handler   = &proc_dointvec,
2063         },
2064 #endif
2065         { .ctl_name = 0 },
2066 };
2067
2068 static ctl_table fs_table[] = {
2069         {
2070                 .ctl_name       = FS_DQSTATS,
2071                 .procname       = "quota",
2072                 .mode           = 0555,
2073                 .child          = fs_dqstats_table,
2074         },
2075         { .ctl_name = 0 },
2076 };
2077
2078 static ctl_table sys_table[] = {
2079         {
2080                 .ctl_name       = CTL_FS,
2081                 .procname       = "fs",
2082                 .mode           = 0555,
2083                 .child          = fs_table,
2084         },
2085         { .ctl_name = 0 },
2086 };
2087
2088 static int __init dquot_init(void)
2089 {
2090         int i;
2091         unsigned long nr_hash, order;
2092
2093         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2094
2095         register_sysctl_table(sys_table);
2096
2097         dquot_cachep = kmem_cache_create("dquot",
2098                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2099                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2100                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2101                         NULL);
2102
2103         order = 0;
2104         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2105         if (!dquot_hash)
2106                 panic("Cannot create dquot hash table");
2107
2108         /* Find power-of-two hlist_heads which can fit into allocation */
2109         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2110         dq_hash_bits = 0;
2111         do {
2112                 dq_hash_bits++;
2113         } while (nr_hash >> dq_hash_bits);
2114         dq_hash_bits--;
2115
2116         nr_hash = 1UL << dq_hash_bits;
2117         dq_hash_mask = nr_hash - 1;
2118         for (i = 0; i < nr_hash; i++)
2119                 INIT_HLIST_HEAD(dquot_hash + i);
2120
2121         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2122                         nr_hash, order, (PAGE_SIZE << order));
2123
2124         register_shrinker(&dqcache_shrinker);
2125
2126 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
2127         if (genl_register_family(&quota_genl_family) != 0)
2128                 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
2129 #endif
2130
2131         return 0;
2132 }
2133 module_init(dquot_init);
2134
2135 EXPORT_SYMBOL(register_quota_format);
2136 EXPORT_SYMBOL(unregister_quota_format);
2137 EXPORT_SYMBOL(dqstats);
2138 EXPORT_SYMBOL(dq_data_lock);
2139 EXPORT_SYMBOL(vfs_quota_on);
2140 EXPORT_SYMBOL(vfs_quota_on_mount);
2141 EXPORT_SYMBOL(vfs_quota_off);
2142 EXPORT_SYMBOL(vfs_quota_sync);
2143 EXPORT_SYMBOL(vfs_get_dqinfo);
2144 EXPORT_SYMBOL(vfs_set_dqinfo);
2145 EXPORT_SYMBOL(vfs_get_dqblk);
2146 EXPORT_SYMBOL(vfs_set_dqblk);
2147 EXPORT_SYMBOL(dquot_commit);
2148 EXPORT_SYMBOL(dquot_commit_info);
2149 EXPORT_SYMBOL(dquot_acquire);
2150 EXPORT_SYMBOL(dquot_release);
2151 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
2152 EXPORT_SYMBOL(dquot_initialize);
2153 EXPORT_SYMBOL(dquot_drop);
2154 EXPORT_SYMBOL(vfs_dq_drop);
2155 EXPORT_SYMBOL(dquot_alloc_space);
2156 EXPORT_SYMBOL(dquot_alloc_inode);
2157 EXPORT_SYMBOL(dquot_free_space);
2158 EXPORT_SYMBOL(dquot_free_inode);
2159 EXPORT_SYMBOL(dquot_transfer);
2160 EXPORT_SYMBOL(vfs_dq_transfer);
2161 EXPORT_SYMBOL(vfs_dq_quota_on_remount);